You're excluding a major component of the process: hypothesis.
Step 1) observe a phenomenon
Step 2) form a hypothesis about the phenomenon
Step 3) design a repeatable and falsifiable experiment to confirm your theory
Step 4) tell people about your findings

This process is exactly the same as superstition formation sans the "repeatable and falsifiable" part. Much science is based on statistical analysis, this is the experiment. Correlation does not prove causation, therefore most hypotheses are just wild ass guesses. No different than a rain dance.